[lld] e32f04c - [ELF] Parse SHT_GNU_verneed and respect versioned undefined symbols in shared objects
Fangrui Song via llvm-commits
llvm-commits at lists.llvm.org
Sat May 23 09:56:39 PDT 2020
Author: Fangrui Song
Date: 2020-05-23T09:55:48-07:00
New Revision: e32f04cdc95224589f30148599c362ba37bae7b6
URL: https://github.com/llvm/llvm-project/commit/e32f04cdc95224589f30148599c362ba37bae7b6
DIFF: https://github.com/llvm/llvm-project/commit/e32f04cdc95224589f30148599c362ba37bae7b6.diff
LOG: [ELF] Parse SHT_GNU_verneed and respect versioned undefined symbols in shared objects
An undefined symbol in a shared object can be versioned, like `f at v1`.
We currently insert `f` as an Undefined into the symbol table, but we
should insert `f at v1` instead.
The string `v1` is inferred from SHT_GNU_versym and SHT_GNU_verneed.
This patch implements the functionality.
Failing to do this can cause two issues:
* If a versioned symbol referenced by a shared object is defined in the
executable, we will fail to export it.
* If a versioned symbol referenced by a shared object in another object
file, --no-allow-shlib-undefined may spuriously report an
"undefined reference to " error. See https://bugs.llvm.org/show_bug.cgi?id=44842
(Linking -lfftw3 -lm on Arch Linux can cause
`undefined reference to __log_finite`)
Reviewed By: grimar
Differential Revision: https://reviews.llvm.org/D80059
Added:
lld/test/ELF/invalid/verneed-shared.yaml
lld/test/ELF/verneed-shared.s
Modified:
lld/ELF/InputFiles.cpp
lld/ELF/InputFiles.h
Removed:
################################################################################
diff --git a/lld/ELF/InputFiles.cpp b/lld/ELF/InputFiles.cpp
index 7a165e28cdc4..9fdd0547ddca 100644
--- a/lld/ELF/InputFiles.cpp
+++ b/lld/ELF/InputFiles.cpp
@@ -1211,6 +1211,42 @@ static std::vector<const void *> parseVerdefs(const uint8_t *base,
return verdefs;
}
+// Parse SHT_GNU_verneed to properly set the name of a versioned undefined
+// symbol. We detect fatal issues which would cause vulnerabilities, but do not
+// implement sophisticated error checking like in llvm-readobj because the value
+// of such diagnostics is low.
+template <typename ELFT>
+std::vector<uint32_t> SharedFile::parseVerneed(const ELFFile<ELFT> &obj,
+ const typename ELFT::Shdr *sec) {
+ if (!sec)
+ return {};
+ std::vector<uint32_t> verneeds;
+ ArrayRef<uint8_t> data = CHECK(obj.getSectionContents(sec), this);
+ const uint8_t *verneedBuf = data.begin();
+ for (unsigned i = 0; i != sec->sh_info; ++i) {
+ if (verneedBuf + sizeof(typename ELFT::Verneed) > data.end() ||
+ uintptr_t(verneedBuf) % sizeof(uint32_t) != 0)
+ fatal(toString(this) + " has an invalid Verneed");
+ auto *vn = reinterpret_cast<const typename ELFT::Verneed *>(verneedBuf);
+ const uint8_t *vernauxBuf = verneedBuf + vn->vn_aux;
+ for (unsigned j = 0; j != vn->vn_cnt; ++j) {
+ if (vernauxBuf + sizeof(typename ELFT::Vernaux) > data.end() ||
+ uintptr_t(vernauxBuf) % sizeof(uint32_t) != 0)
+ fatal(toString(this) + " has an invalid Vernaux");
+ auto *aux = reinterpret_cast<const typename ELFT::Vernaux *>(vernauxBuf);
+ if (aux->vna_name >= this->stringTable.size())
+ fatal(toString(this) + " has a Vernaux with an invalid vna_name");
+ uint16_t version = aux->vna_other & VERSYM_VERSION;
+ if (version >= verneeds.size())
+ verneeds.resize(version + 1);
+ verneeds[version] = aux->vna_name;
+ vernauxBuf += aux->vna_next;
+ }
+ verneedBuf += vn->vn_next;
+ }
+ return verneeds;
+}
+
// We do not usually care about alignments of data in shared object
// files because the loader takes care of it. However, if we promote a
// DSO symbol to point to .bss due to copy relocation, we need to keep
@@ -1254,6 +1290,7 @@ template <class ELFT> void SharedFile::parse() {
const Elf_Shdr *versymSec = nullptr;
const Elf_Shdr *verdefSec = nullptr;
+ const Elf_Shdr *verneedSec = nullptr;
// Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d.
for (const Elf_Shdr &sec : sections) {
@@ -1270,6 +1307,9 @@ template <class ELFT> void SharedFile::parse() {
case SHT_GNU_verdef:
verdefSec = &sec;
break;
+ case SHT_GNU_verneed:
+ verneedSec = &sec;
+ break;
}
}
@@ -1309,12 +1349,13 @@ template <class ELFT> void SharedFile::parse() {
sharedFiles.push_back(this);
verdefs = parseVerdefs<ELFT>(obj.base(), verdefSec);
+ std::vector<uint32_t> verneeds = parseVerneed<ELFT>(obj, verneedSec);
// Parse ".gnu.version" section which is a parallel array for the symbol
// table. If a given file doesn't have a ".gnu.version" section, we use
// VER_NDX_GLOBAL.
size_t size = numELFSyms - firstGlobal;
- std::vector<uint32_t> versyms(size, VER_NDX_GLOBAL);
+ std::vector<uint16_t> versyms(size, VER_NDX_GLOBAL);
if (versymSec) {
ArrayRef<Elf_Versym> versym =
CHECK(obj.template getSectionContentsAsArray<Elf_Versym>(versymSec),
@@ -1345,7 +1386,22 @@ template <class ELFT> void SharedFile::parse() {
continue;
}
+ uint16_t idx = versyms[i] & ~VERSYM_HIDDEN;
if (sym.isUndefined()) {
+ // For unversioned undefined symbols, VER_NDX_GLOBAL makes more sense but
+ // as of binutils 2.34, GNU ld produces VER_NDX_LOCAL.
+ if (idx != VER_NDX_LOCAL && idx != VER_NDX_GLOBAL) {
+ if (idx >= verneeds.size()) {
+ error("corrupt input file: version need index " + Twine(idx) +
+ " for symbol " + name + " is out of bounds\n>>> defined in " +
+ toString(this));
+ continue;
+ }
+ StringRef verName = this->stringTable.data() + verneeds[idx];
+ versionedNameBuffer.clear();
+ name =
+ saver.save((name + "@" + verName).toStringRef(versionedNameBuffer));
+ }
Symbol *s = symtab->addSymbol(
Undefined{this, name, sym.getBinding(), sym.st_other, sym.getType()});
s->exportDynamic = true;
@@ -1355,7 +1411,6 @@ template <class ELFT> void SharedFile::parse() {
// MIPS BFD linker puts _gp_disp symbol into DSO files and incorrectly
// assigns VER_NDX_LOCAL to this section global symbol. Here is a
// workaround for this bug.
- uint32_t idx = versyms[i] & ~VERSYM_HIDDEN;
if (config->emachine == EM_MIPS && idx == VER_NDX_LOCAL &&
name == "_gp_disp")
continue;
diff --git a/lld/ELF/InputFiles.h b/lld/ELF/InputFiles.h
index 036b075931e6..51882e0c9647 100644
--- a/lld/ELF/InputFiles.h
+++ b/lld/ELF/InputFiles.h
@@ -370,6 +370,11 @@ class SharedFile : public ELFFileBase {
// Used for --as-needed
bool isNeeded;
+
+private:
+ template <typename ELFT>
+ std::vector<uint32_t> parseVerneed(const llvm::object::ELFFile<ELFT> &obj,
+ const typename ELFT::Shdr *sec);
};
class BinaryFile : public InputFile {
diff --git a/lld/test/ELF/invalid/verneed-shared.yaml b/lld/test/ELF/invalid/verneed-shared.yaml
new file mode 100644
index 000000000000..18315fe8a2df
--- /dev/null
+++ b/lld/test/ELF/invalid/verneed-shared.yaml
@@ -0,0 +1,79 @@
+## REQUIRES: x86
+## Test that we can parse SHT_GNU_verneed in a shared object and report certain errors.
+
+# RUN: echo '.globl _start; _start:' | llvm-mc -filetype=obj -triple=x86_64 - -o %t.o
+
+## sh_offset(SHT_GNU_verneed) is out of bounds.
+# RUN: yaml2obj --docnum=1 %s -o %t1.so
+# RUN: not ld.lld %t.o %t1.so -o /dev/null 2>&1 | FileCheck --check-prefix=SHOFFSET %s
+# SHOFFSET: error: {{.*}}.so: section [index 1] has a sh_offset (0xffffffff) + sh_size (0x0) that is greater than the file size (0x228)
+--- !ELF
+FileHeader:
+ Class: ELFCLASS64
+ Data: ELFDATA2LSB
+ Type: ET_DYN
+ Machine: EM_X86_64
+Sections:
+ - Name: .gnu.version_r
+ Type: SHT_GNU_verneed
+ Flags: [ SHF_ALLOC ]
+ ShOffset: 0xFFFFFFFF
+
+## A Verneed entry is misaligned (not a multiple of 4).
+# RUN: yaml2obj --docnum=2 %s -o %t2.so
+# RUN: not ld.lld %t.o %t2.so -o /dev/null 2>&1 | FileCheck --check-prefix=VN-MISALIGNED %s
+# VN-MISALIGNED: {{.*}}.so has an invalid Verneed
+--- !ELF
+FileHeader:
+ Class: ELFCLASS64
+ Data: ELFDATA2LSB
+ Type: ET_DYN
+ Machine: EM_X86_64
+Sections:
+ - Type: Fill
+ Size: 0x1
+ - Name: .gnu.version_r
+ Type: SHT_GNU_verneed
+ Flags: [ SHF_ALLOC ]
+ Info: 1
+ Link: .dynstr
+ Dependencies:
+ - Version: 1
+ File: foo
+ Entries:
+ - Name: 'foo'
+ Hash: 0
+ Flags: 0
+ Other: 0
+DynamicSymbols:
+ - Name: foo
+
+## vn_aux points to a place outside of the file.
+# RUN: yaml2obj --docnum=3 -D VERNEED=0100010001000000040200000000000000000000 %s -o %t3.so
+# RUN: not ld.lld %t.o %t3.so -o /dev/null 2>&1 | FileCheck --check-prefix=AUX-OOB %s
+# AUX-OOB: {{.*}}.so has an invalid Vernaux
+--- !ELF
+FileHeader:
+ Class: ELFCLASS64
+ Data: ELFDATA2LSB
+ Type: ET_DYN
+ Machine: EM_X86_64
+Sections:
+ - Name: .gnu.version_r
+ Type: SHT_GNU_verneed
+ Flags: [ SHF_ALLOC ]
+ Info: 1
+ Link: .dynstr
+ Content: "[[VERNEED]]"
+DynamicSymbols:
+ - Name: foo
+
+## vn_aux is misaligned.
+# RUN: yaml2obj --docnum=3 -D VERNEED=0100010001000000110000000000000000000000 %s -o %t4.so
+# RUN: not ld.lld %t.o %t4.so -o /dev/null 2>&1 | FileCheck --check-prefix=AUX-MISALIGNED %s
+# AUX-MISALIGNED: {{.*}}.so has an invalid Vernaux
+
+## vna_name is out of bounds.
+# RUN: yaml2obj --docnum=3 -D VERNEED=010001000000000010000000000000009107000000000000ff00000000000000 %s -o %t5.so
+# RUN: not ld.lld %t.o %t5.so -o /dev/null 2>&1 | FileCheck --check-prefix=NAME-OOB %s
+# NAME-OOB: {{.*}}.so has a Vernaux with an invalid vna_name
diff --git a/lld/test/ELF/verneed-shared.s b/lld/test/ELF/verneed-shared.s
new file mode 100644
index 000000000000..218f43c2545e
--- /dev/null
+++ b/lld/test/ELF/verneed-shared.s
@@ -0,0 +1,37 @@
+# REQUIRES: x86
+# RUN: echo 'v1 { f; }; v2 { g; };' > %t.ver
+# RUN: llvm-mc -filetype=obj -triple=x86_64 %s -o %t.o
+# RUN: ld.lld -shared --version-script %t.ver %t.o -o %t.so
+
+# RUN: ld.lld --version-script %t.ver %t.o %t.so -o /dev/null -y f at v1 | \
+# RUN: FileCheck --check-prefix=TRACE %s --implicit-check-not=f at v1
+
+## TRACE: {{.*}}.o: definition of f at v1
+## TRACE-NEXT: {{.*}}.so: shared definition of f at v1
+
+# RUN: echo '.symver f,f at v1; .symver g,g at v2; call f; call g' | \
+# RUN: llvm-mc -filetype=obj -triple=x86_64 - -o %t1.o
+# RUN: ld.lld -shared %t1.o %t.so -o %t1.so
+
+## Test that we can parse SHT_GNU_verneed to know that the undefined symbols in
+## %t1.so are called 'f at v1' and 'g at v2', which can be satisfied by the executable.
+## We will thus export the symbols.
+# RUN: ld.lld -pie %t.o %t1.so -o %t
+# RUN: llvm-nm -D %t | FileCheck --check-prefix=NM %s
+
+# NM: T f
+# NM: T g
+
+## The default is --no-allow-shlib-undefined.
+## Don't error because undefined symbols in %t1.so are satisfied by %t.so
+# RUN: llvm-mc -filetype=obj -triple=x86_64 /dev/null -o %t2.o
+# RUN: ld.lld %t2.o %t1.so %t.so -y f at v1 -o /dev/null | FileCheck %s
+
+# CHECK: {{.*}}1.so: reference to f at v1
+# CHECK-NEXT: {{.*}}.so: shared definition of f at v1
+
+.globl f_v1, g_v2
+.symver f_v1,f at v1
+.symver g_v2,g at v2
+f_v1:
+g_v2:
More information about the llvm-commits
mailing list